SemaGBA: How Semaglutide Regulates Appetite and Weight

Abstract

Study at a Glance

Study Snapshot

All baseline values and model predictions derived from Kennis & van Riel 2026, Table 1.

Study Facts Table

What the Researchers Actually Did

Kennis and van Riel developed SemaGBA, a system dynamics model implemented in Julia, to simulate how semaglutide modulates a network of 14 interconnected metabolic and neural variables. The model uses differential equations to represent how each variable changes over time as a daily average. Dimensional variables — those with clinical units such as blood glucose (mg/dL) and body weight (kg) — are directly measurable. Dimensionless variables — insulin sensitivity, leptin sensitivity, beta-cell function, glucotoxicity, lipotoxicity, and the three neural activity variables — are normalized relative to a healthy baseline and represent processes not routinely captured in clinical settings.

Model construction followed a staged workflow. A reduced version omitting neural variables was calibrated iteratively against phase 3 trial data from SUSTAIN 1 (type 2 diabetes) and STEP 1, 3, and 4 (obesity). Once metabolic outputs were consistent with observed outcomes, the model was extended to include POMC, AgRP, and dopamine neuron activity, which were constrained to influence only net energy intake and each other, with a maximum deviation from the reduced model of 2.8% in the obesity simulation. Semaglutide’s pharmacokinetics were described using a first-order, single-compartment model with subcutaneous doses added every seven days beginning at day 7 of simulation, following the standard clinical escalation schedule (0.25, 0.5, 1.0, 1.7, and 2.4 mg). The drug was assumed to exert the same physiological effects as endogenous GLP-1.

Key Findings: Primary Outcomes

• At 0.5 mg semaglutide over 30 weeks in the type 2 diabetes simulation, fasting blood glucose decreased by 38.0 mg/dL (observed: 41.0 mg/dL) and body weight decreased by 3.2 kg (observed: 3.8 kg).
• At 1.0 mg semaglutide over 30 weeks, glucose decreased by 45.4 mg/dL (observed: 44.0 mg/dL) and body weight by 4.1 kg (observed: 4.7 kg).
• At 2.4 mg semaglutide over 68 weeks in the obesity simulation, body weight decreased by 15.1%, compared with observed values of 14.9% (STEP 1), 16.0% (STEP 3), and 17.1% (STEP 4).
• In the type 2 diabetes simulation, semaglutide reduced net energy intake by 5.5%, with beta-cell function rising from 0.50 to 0.89 and glucotoxicity falling from 0.50 to 0.37.
• In the obesity simulation, net energy intake fell by 26.0%, lipotoxicity decreased from 0.40 to 0.27, and beta-cell function improved from 0.80 to 1.4. Insulin secretion transiently rose to a maximum of 17.5 µU/mL before declining to 11.4 µU/mL as blood glucose normalized toward 90–95 mg/dL, consistent with the model’s representation of glucose-dependent insulin secretion.

Key Findings: Secondary Outcomes and Subgroup Analyses

• Prediabetes simulation: In a healthy individual modeled to chronically consume 2,700 kcal/day, untreated overeating produced obesity (BMI 30.4 kg/m²) within 27 months and type 2 diabetes (blood glucose ≥125 mg/dL) within 31 months. Initiating 1.0 mg semaglutide at day 245 (prediabetic phase) maintained blood glucose at approximately 101.0 mg/dL, keeping glucotoxicity at 0.006 versus 0.23 in the untreated trajectory and preserving beta-cell function at 1.57 versus 0.88 untreated. Because net energy intake was held constant in this simulation, weight-mediated variables (leptin, lipotoxicity) were unaffected.
• Neural variable simulation: POMC neuron activity increased in both diabetes and obesity simulations, driven primarily by semaglutide dosage escalation and rising insulin levels. AgRP activity declined initially due to insulin inhibition, then rose modestly as leptin levels fell. Dopamine neuron activity decreased in both simulations due to insulin and semaglutide inhibition, with a modest rebound late in each simulation as leptin declined. Maximum deviations in net energy intake between the reduced and extended models were 1.4% (diabetes) and 2.8% (obesity).
• The model predicted that declining leptin during treatment generates counter-regulatory appetite stimulation via AgRP, which the authors propose may attenuate semaglutide’s long-term effectiveness.
• The model also suggested that dopamine neuron activity responds earliest to semaglutide, implying that patients may experience reduced food reward before homeostatic satiety signals via POMC and AgRP pathways become dominant.

Adverse Events and Safety Profile

Statistical Approach and Rigor

SemaGBA is a deterministic system dynamics model; it produces single-trajectory outputs for defined initial conditions rather than probabilistic estimates with confidence intervals. Effect function parameters were calibrated iteratively by comparing model outputs to aggregate clinical trial means, not to individual patient data. No formal goodness-of-fit statistics, residual analyses, or sensitivity analyses are reported in the main text (supplementary materials contain additional detail). The model does not incorporate stochastic variability, so it cannot capture the distribution of individual responses seen in clinical practice. Validation was restricted to aggregate endpoints from five trials over a maximum of 68 weeks, leaving longer-term predictions unvalidated. The neural variable outputs are not calibrated to any human clinical data; their plausibility depends entirely on the physiological assumptions encoded in the effect functions.

Clinical Takeaway

SemaGBA provides a mechanistically grounded, computationally tractable framework for understanding why semaglutide produces the clinical outcomes observed in pivotal trials. The model’s close reproduction of blood glucose and weight loss endpoints across five phase 3 datasets is reassuring regarding its metabolic architecture. Clinicians should understand that the neural outputs — the AgRP, POMC, and dopamine trajectories — are hypothesis-generating constructs, not validated predictions. The prediabetes simulation is particularly speculative, as net energy intake was held constant (a condition that does not reflect clinical reality), producing pharmacologically unrealistic fasting insulin values above 90 µU/mL. The model’s greatest near-term utility is conceptual: it illustrates how semaglutide’s glucose-dependent insulin secretion, beta-cell preservation, and appetite suppression form an integrated, self-reinforcing system rather than isolated mechanisms.

Why This Matters Clinically

The mechanisms by which semaglutide produces sustained weight loss and glycemic improvement remain incompletely understood even after years of widespread clinical use. Most prior computational models of semaglutide described dose-response relationships for weight or glucose without integrating the feedback loops that produce those outcomes, and experimental neural data come from animal studies that have not been linked to metabolic endpoints or clinical measures. SemaGBA is the first published model to couple metabolic dynamics — beta-cell function, glucotoxicity, lipotoxicity, insulin and leptin sensitivity — with simulated neural activity in AgRP, POMC, and dopamine pathways, and to ground that coupled model in phase 3 human trial data. This matters because it provides a testable, mechanistically explicit framework that can guide the design of future human studies and, potentially, inform dose optimization or the identification of biomarkers that predict treatment response.

CED Clinical Relevance

At CED Clinic, patients receiving semaglutide often ask why the medication reduces their appetite, whether the effect is durable, and why some individuals plateau. SemaGBA offers a rigorous framework for answering those questions. The model’s depiction of declining leptin levels generating counter-regulatory appetite signals through AgRP neurons — partially offsetting semaglutide’s anorexigenic effect — aligns with what clinicians observe when appetite suppression attenuates over months of treatment. The prediabetes scenario reinforces the clinical case for early intervention before glucotoxic damage accumulates, a principle already central to CED’s approach. Importantly, the model’s limitations remind us that the neural architecture remains speculative: we should not overstate certainty about specific neural mechanisms to patients, even as the metabolic framework is well-grounded.

Fits What We Already Know

The SemaGBA model is grounded in well-established physiology. GLP-1’s stimulation of glucose-dependent insulin secretion, its enhancement of beta-cell mass through neogenesis and apoptosis inhibition, and its role in slowing gastric emptying are cited from prescribing information for Ozempic and Wegovy. The model’s depiction of lipotoxicity impairing beta-cell function and contributing to insulin resistance draws on the glucotoxicity-lipotoxicity convergence framework described by Poitout and Robertson (2002). The POMC and AgRP neuronal circuitry is consistent with Varela and Horvath (2012) and Wu and Zheng (2018). Experimental data confirming semaglutide’s effects on POMC, AgRP, and dopamine neurons in animal models come from multiple cited studies, including Kooij et al. (2024) and Zhu et al. (2025). The prediabetes intervention findings align directionally with the meta-analysis by Salamah et al. (2024) on GLP-1 receptor agonists in prediabetes, though semaglutide remains without FDA approval for this indication as of this publication.

What This Study Teaches Us

Semaglutide’s clinical effects arise from an interconnected system, not a single mechanism. Weight loss improves lipotoxicity, which in turn restores leptin and insulin sensitivity and protects beta-cell function. Semaglutide simultaneously reinforces this trajectory by directly enhancing beta-cell function and sustaining GLP-1 receptor activation at levels that prevent the compensatory appetite increases that would otherwise accompany caloric restriction and leptin decline. The neural layer adds a plausible explanation for early appetite suppression through dopamine pathways and sustained satiety through POMC activation, with AgRP activity tracking a counter-regulatory arc that may explain why some patients experience appetite attenuation over time. All of that neural narrative is hypothesis, not established fact — but it is a well-structured hypothesis built on a metabolic framework that does match clinical data.

What It Does Not Show

• The neural variable outputs (AgRP, POMC, dopamine) are not validated against any human clinical or imaging data. They represent one internally consistent solution, not the unique or correct mechanistic answer.
• The model does not capture individual patient variability; it produces single-trajectory outputs for population archetypes.
• The prediabetes simulation holds net energy intake constant, precluding weight loss and generating physiologically unrealistic fasting insulin levels (93.2 µU/mL), which limits its translational value.
• Glucagon dynamics — suppressed by semaglutide and clinically relevant to glucose regulation — are omitted.
• Fat depot location and distribution are not distinguished, despite their differential metabolic significance.
• Physiological adaptations to sustained weight loss (e.g., reduced resting metabolic rate) are not incorporated.
• Social, behavioral, and environmental drivers of food intake are excluded by design.
• Long-term predictions beyond 68 weeks (the maximum validation window) are extrapolations without clinical grounding.
• Treatment discontinuation, dose adjustment for tolerability, and non-responders are not modeled.

Fits the Broader Conversation

The field

Physician-Led, Whole-Person Care

A doctor who takes the time to truly understand you.

Personal care that starts with listening and is guided by experience and ingenuity.

Health, Longevity, Wellness

One-on-One Cannabis Guidance

Metabolic Balance

Leave a Message
Metabolic Care
Medical Consulting
Cannabis Care

• Newsletter Signup Form

  Enter your email address to subscribe to our newsletter.
• Email Address*
  example@example.com
• Subscribe
• Should be Empty: